Validation of SeaWiFS chlorophyll a concentrations in the Southern Ocean using insitu HPLC measurements

Marina Marrari*, Chuanmin Hu and Kendra Daly

College of Marine Science

University of SouthFlorida

140 Seventh Avenue South

St. Petersburg, FL33701, USA

* Corresponding author: ; Tel: 727-553-1207; Fax: 727-553-1186

Abstract

Surface chlorophyll a concentrations (Ca, mg m-3) in the Southern Ocean estimated from SeaWiFS satellite data have been reported in the literature to be significantly lower than those measured from in situ water samples using fluorometric methods. However, we found that high-resolution (~ 1 km2per pixel) daily SeaWiFS Ca (CaSWF)data (SeaDAS4.8, OC4v4 algorithm) was an accurate measure of in situCaduring January-February of 1998-2002 based onif concurrent in situ data frommeasured byboth fluorometric (CaFluor) and HPLC (CaHPLC) were used as groundtruth instead of fluorometric (CaFluor) measurements. was used as the ground truth collected during January-February of 1998-2002. Our analyses indicate thatCaFluor is 2.482.23 (n=647) times greater thanCaHPLCbetween 0.05 and 1.5 mg m-3 and that thepercentage overestimation of in situCa by fluorometric measurements increases with decreasing concentrations. The ratio of CaSWF/CaHPLC is 1.12  0.91 (n=96), whereas the ratio of CaSWF/CaFluor is0.55 0.63 (n=307). Furthermore, there is no significant bias in CaSWF (12% and -0.07 in linear and log-transformed Ca, respectively) whenCaHPLCis used as groundtruth instead of CaFluor. The high CaFluor/CaHPLC ratiomay be attributed to the relatively low concentrations of chlorophyll b (Cb/Ca = 0.0230.034, n=486) and relatively high concentrations of chlorophyll c (Cc/Ca = 0.250.59, n=486) in the phytoplankton pigment composition when compared to values from other regions. Because more than 90% of the waters in the study area, as well as in the entire Southern Ocean (south of 60oS), have CaSWFbetween 0.05 and 1.5 mg m-3, we consider that the SeaWiFS performance of Ca retrieval is satisfactory and there is no need to further develop a “regional” bio-optical algorithm to account for the previous SeaWiFS “underestimation”.

Keywords: Remote Sensing, Ocean Color, Algorithm, Chlorophyll, HPLC, Fluorometric, Southern Ocean.

Introduction

Since the launch of the Sea-viewing Wide Field-of-view Sensor (SeaWiFS, McClain et al., 1998) onboard the Orbview-II satellite in August 1997, ocean color data products, in particular concentrations of chlorophylla(Ca, mg m-3)in the surface ocean, have been used to investigate a wide variety of fundamentaltopics including ocean primary productivity, biogeochemistry, coastal upwelling, eutrophication, and harmful algal blooms(e.g., Hu et al., 2005; Muller-Karger et al., 2004). Other ocean color missions, such as the ongoing MODerate-resolution Imaging Spectroradiometer (MODIS, Esaias et al., 1998; Terra satellite for morning pass since 1999 and Aqua satellite for afternoon pass since 2002) or the future National Polar-Orbiting Operational Environmental Satellite System (NPOESS), assure the continuity ofremotely sensedocean color in assessing the long-term global change in several key environmental parameters, including Ca. Quantitative use of ocean color data products requires a high level of accuracy. Validation efforts during algorithm development show that errors in the Ca data products after logarithmic transformation are about 0.2 or less (O’Reilly et al., 2000), which corresponds to roughly 50% root mean square (RMS) relative error. Previous studies also show that in most ocean basins, Ca errors are about 0.3 (Gregg and Casey, 2004) although in regions such as the Southern Ocean, reported errors are significantly larger.

The Southern Ocean (SO) was defined by the International Hydrographic Organization in 2000 to encompass waters between the northern coast of Antarctica and 60°S. Oceanographers, however, traditionally have defined the northern limit of the SO as the Subtropical Front (at approximately 40º S) (Orsi et al., 1995). Typical chlorophyll concentrations in the SO range between 0.05 and 1.5 mg m-3 (Arrigo et al., 1998; El-Sayed, 2005). It is believed that the interaction of light and deep mixing, iron, and grazing limit phytoplankton growth throughout the SO, in addition to low silicate concentrations which can limit diatom production north of the Polar Front (Moline and Prézelin, 1996; Daly et al., 2001; Boyd, 2002). However, elevated chlorophyll concentrations (1 to > 30 mg m-3) are characteristic of many regions, including continental shelf and ice edge areas (Holm-Hansen et al., 1989; Moore and Abbott, 2000; El-Sayed, 2005), and even values of up to 190 mg m-3 have been reported (El-Sayed, 1971) . The Antarctic Peninsula region, in particular, supports large concentrations of phytoplankton, zooplankton, seabirds, seals and whales, and is considered one of the most productive areas of the Southern Ocean, for reasons that are not fully understood (Deibel and Daly, in press).

Several studies have relied on ocean color datato investigate phytoplankton spatial patterns (Moore & Abbott, 2000; Holm Hansen et al., 2004), interannual variability during summer (Smith et al., 1998; Korb et al., 2004) and primary productivity (Dierssen et al., 2000; Smith et al., 2001) west of the Antarctic Peninsula and in the adjoining Scotia Sea. These studies usedin situCadetermined from water samples using fluorometric methods(CaFluor) to validate monthly/weekly averages of SeaWiFSCa(CaSWF)data product at ~ 9 x 9 km2or ~ 4 x 4 km2resolution and concluded that in the Southern Ocean, CaSWFvalues are significantly lower than those estimated from in situ water samples. For example, Dierssen and Smith (2000) appliedin situbio-optical datameasured between 1991 and 1998 to the OC2v2 algorithm to test its applicability to the standard SeaWiFS algorithm to test its applicability in the Southern Ocean, and concluded that, west of the Antarctic Peninsula in the Southern Ocean. They concluded that , Caderived from the OC2v2 algorithm using in situ reflectance was 60% lower than in situCa(between 1991 and 1998 (Ca between 0.7 and 43 mg m-3, median ~ 1 mg m-3). Korb et al. (2004) reported that CaSWFvalues were only 87% of CaFluorfor concentrations lower than 1 mg m-3 and only 30% for concentrations above 5 mg m-3in the South Georgia area (54.5º S, 37º W). In addition, Moore et al. (1999) found a strong linear relationship between CaSWF and CaFluor(R2 = 0.72, n= 84) in the RossSea, although they noted that SeaWiFS tended to underestimate Cavalues between 0.1 and 1.5 mg m-3.

The previous validation methods may present several limitations. First, in situ samplesare point measurements while satellite pixels cover a larger area (up to 9x9 km2). Patchiness within a pixel will affect the comparison of results between areas and over time (e.g., Hu et al., 2004). Second, the in situ and satellite measurements are not strictly concurrent and the time differences can be large (up to a month). Finally, and most importantly, previous validation studies used in situ Cafrom fluorometric measurements, while it is now widely recognized that High Performance Liquid Chromatography (HPLC) may yield more accurate results in determining Cafrom water samples. Fluorometric methods may result in biased results, particularly in the presence of certain accessory pigments (Lorenzen, 1981; Welschmeyer, 1994).

In a study that included three different areas of the world’s oceans, Trees et al. (1985) reported that errors in the CaFluorranged between -68 and 53% with a meanof 39%. In addition, Bianchi et al. (1995) found that CaFluorin the northern Gulf of Mexico was approximately 30% lower than CaHPLC, except in near coastal areas. It is believed that the presence of significant amounts of chlorophyll b (Cb), characteristic of chlorophytes, prochlorophytes and cryptophytes, causes fluorometric techniques to underestimate Ca. On the other hand, high concentrations of chlorophyll c(Cc),typically found in diatoms, dinoflagellates, prasinophytes and haptophytes, lead to an overestimation of Cawith respect to fluorometric measurements. The fluorescence emission spectra of degradation products (phaeopigments) of Caand Cb overlap considerably, causing an overestimation of Ca phaeopigments and, thus, an underestimation of Ca. On the other hand, Caand Cc have partially overlapping fluorescence spectra, causing an overestimation of Caand subsequent underestimation of phaeopigments a (Gibbs, 1979; Jeffrey et al., 1997). The filters used in the standard fluorometric method (Lorenzen, 1981) cannot effectively discriminate between Ca, Cb, Cc and their degradation products; thus, depending on the type of phytoplankton present and their associated pigments, Camay be overestimated or underestimated by fluorometric methods.

Herein, we use concurrent HPLC and fluorometric data collected between 1998 and 2002in waters west of the Antarctic Peninsula, as well as high-resolution SeaWiFS data, to re-examine whether SeaWiFS Cais underestimatedin the Southern Ocean as reported in previous studies. We also discuss possible explanations for the observed results and investigate the effects of different accessory pigments on Caestimations.

Methods

SeaWiFS daily Level 2 data between December 1997 and December 2004 were obtained from NASA Goddard Space Flight Center( These data were derived from the high-resolution (~ 1 km/pixel near nadir) Level 1data collected by all HRPT ground stations, as well as occasional onboard recording over the area using the most current algorithms and software package (SeaDAS4.8). A total of 6606 data files were obtained and mapped to a rectangular projection with approximately 1 km2/pixel for the area between 45-75˚S and 50-80˚W west of the Antarctic Peninsula (Fig. 1). The data product used in this study is the surface Caestimated with the OC4v4 empirical algorithm (O’Reilly, 2000):

Ca = 10 0.366 - 3.067R + 1.93R^2 + 0.649R^3 - 1.532R^4 (1)

where R=log10[(max((Rrs443, Rrs490,0 Rrs510))/Rrs555)] and Rrs is the remote sensing reflectance, a data product after atmospheric correction.

Chlorophyll fluorescence and HPLC pigment data were collected and analyzed by Drs Raymond Smith (University of California Santa Barbara) and Maria Vernet (University of California San Diego) as part of the Palmer Long Term Ecological Research (LTER) program during cruises west of the Antarctic Peninsula (see for detailed methods). The location of the LTER chlorophyll sampling stations between 1998 and 2002 are shown in Figure 2. Most of the samples were collected within the 2000 m isobath, although two transects were conducted across Drake Passage in January-February 1999 and 2000 to measure CaFluor. At each station, water column samples were collected at discrete depths for both fluorometric and HPLC measurements. Ca,Cb and Ccwere obtained by HPLC from samples collected at fixed stationsduring January-February 1998 and 1999 following the methods of Wright et al. (1991), and during January-February 2000 and 2001 following the methods of Zapata et al. (2000). Caand phaeopigment concentrations also were obtained by fluorometric methods by measuring total fluorescence and subtracting phaeopigments after acidification from samples collected during January-February 1998, 1999, 2000, 2001 and 2002 following Smith et al. (1981, 1995, 1996). Welschmeyer’s (1994) method, which effectively measures fluorescence from Caonly and reduces interference from Cbor its phaeo-derivatives, was not applied (M. Vernet, pers. comm.).

Because the signal detected by the satellite sensor is an optically-weighted function of signals at all depths (up to 50-60 m for clear waters), we used the method of Gordon (1992) to calculate a depth-weighted chlorophyll concentration, <C>, to compare with satellite estimates:

(2)

where and z is the depth. K is the diffuse attenuation coefficient that is approximated by K (z) 0.121 C(z)0.428 (Morel, 1988). The integration was from 0 to 50m and included 5 or 6 vertical samples at most stations, although in some cases only 3 - 4 samples were available for the calculations. A total of 189 HPLC and 775 fluorometric Ca values were used in our analyses. Because the weighting function, g(z),exponentially decreasesexponentially with increasing depth, <C> is not significantly very different from the surface value, at least for fluorometric Ca(ratio = 1.02±0.15, p = 0.8415 for fluorometric Ca). For the HPLC samples, method, the differences between <C> and surface Ca are significant for the HPLC samples (ratio = 1.05 ± 0.99, p = 0.022). The daily, high-resolution SeaWiFS Cadata were queried to compare with the in situ data in the following manner. To reduce errors caused by digitization and random noise, for each in situ data point, all valid satellite data from a 5x5 pixel box covering the in situ location (except those cloud and land adjacent pixels) were used to compute the median value (Hu et al., 2001). A rigorous comparison between satellite and in situ data should limit the time difference between the two measurements to within 2-3 hours. Due to extended cloud coverage and the occasional presence of sea ice, however, only a small number of HPLC data points were obtained under such rigorous criteria, leading to statistically meaningless results. Therefore, the time difference between satellite and in situ measurements was relaxed to  3 days.

Estimating uncertainty in a satellite-derived parameter with log-normal distribution is not trivial, as discussed in Campbell (submitted). Here, tTwo estimates were used to assess the differences between the in situ and satellite-derived datatwo datasets. First, the root mean square (RMS) and the mean difference (bias) in percentage were defined as:

(3)

where Sis satellite data, Iisin situ data, and n is the number of data pairs. For a normally distributed x,RMS should equal the standard deviation. Further, because the natural distribution of Ca is lognormal (Campbell, 1995), error estimates were also made on the logarithmically transformed (base 10) data:

(4)

These error estimates have been used in recent publications to describe the performance of the ocean color algorithms (O’Reilly et al., 2000) and to validate SeaWiFS global and regional estimates of Ca(Darecki and Stramski, 2004; Gregg and Casey, 2004; Zhang et al., 2006). Note that these latter error estimates cannot be expressed as percentages because they are logarithmically transformed (Campbell, submitted).

Results

Typical CaFluorand CaHPLCdistributions during austral summer arepresented for January-February 1999 (Fig.3). In all years, CaFluorranged from 0.052 to 27.6 mg m-3, with a median of 0.86 mg m-3. CaHPLCwas typically lower and ranged from 0.017 to 14.6 mg m-3 with a median of 1.04 mg m-3. In general, the lowest Cavalues (<0.1 mg m-3) were consistently found offshelf in Drake Passage. Elevated Cavalues (>1 mg m-3) were detected throughout the continental shelf, with the highest values (>10 mg m-3) always observed in MargueriteBay.

A total of 96 CaSWF-CaHPLCmatching pairs and 307 CaSWF-CaFluormatching pairs were obtained using the method described above. Table 1 lists the statistics of these comparisons. In general, CaSWFis significantly lower than CaFluor(Fig.4), with a ratio of 0.55  0.63 between the two (Table 1). The inverse ratio, i.e., the ratio of CaFluor/CaSWF, is 2.73  2.19, consistent with previous observations in the Southern Ocean where CaFluor was used to validate CaSWFand the same pattern of underestimation was observed(Moore et al., 1999; Dierssen and Smith, 2000; Korb et al., 2004). In contrast, CaHPLCshowed a more satisfactory agreement with CaSWFover a wide dynamic range (0.1 – 4 mg m-3) (Fig.4). Themean ratio of CaSWF/CaHPLC is close to 1 (i.e. 1.12), in contrast to the lower ratio of 0.55 for CaSWF/CaFluor.

Although the RMS errors (in natural or log-transformed forms) for the two comparisons are comparable (Table 1), CaHPLCis nearly equally scattered around the 1:1 line (Fig. 4), suggesting that the bias errors in CaSWF/CaHPLC are significantly smaller than those in CaSWF/CaFluor. Clearly, the agreement between CaSWF and CaHPLC is much improved over that between CaSWF and CaFluor.

Similar results were also obtained from the algorithm perspective. By using the spectral remote sensing reflectance satellite data (Rrs)derived from the satellite measurements (Fig. 5), the OC4v4 algorithm yielded comparable results to those obtained from HPLC measurements. In contrast, CaFluorvalues are significantly higher than those predicted by the OC4v4 algorithm for the entire range considered.

Are these results representative of the entire Southern Ocean? Due to cloud cover, satellite data werenot available for all pixels every day. This reduced the number of CaSWF data points, whichresulted in a limited number of matchingpairs for comparingsatellite andin situdata (307 for fluorometric and 96 for HPLC). However, the in situ data itself comprised a much larger dataset that included 832 concurrent fluorometric and HPLC measurements. When this in situ dataset was used tocompareCaFluorand CaHPLC,similar results were obtained, i.e., the ratio of CaFluor/CaHPLCis 2.433.37 (Fig.6). Clearly, tThe ratio of CaFluor/CaHPLC is not uniform, but appears to decrease with increasing concentrations (Table 2),. aAlthough for Indeed, forCaHPLC<0.05 mg m-3 and CaHPLC >3.0 mg m-3, the statistical results may not be reliable because of the few matching pairs available and the scatter of the data (Fig. 6). For CaHPLCbetween 1.5 and 3.0 mg m-3, the bias is insignificant small (15%) and the ratio of CaFluor/CaHPLC is close to unity (1.150.73). For CaHPLC bBetween 0.05 and 1.5 mg m-3, however, there is clearly an overestimation in CaFluoris much higher than CaHPLC (CaFluor/CaHPLC = 2.482.23, n=647), and this difference is believed to be due to errors in the CaFluor measurements as described above.. Because most (>90%) of the waters in the Southern Ocean have surface CaSWF values between 0.05 and 1.5 mg m-3 (Fig.7), this assessment can be generalized and applied to most regions.

Discussion

Although HPLC has been recommended as the most reliable method to determine Ca(e.g. Trees et al., 1985), most cruise surveys still use the fluorometric method because it is faster, requires less technical expertise and is less expensive than HPLC. The Ca data originally used in the development of the OC4v4 algorithm (O’Reilly et al., 2000) included 2,853 in situ measurements from a variety of oceanic environments (but not the Southern Ocean), of which 72% were fluorometric and 28% were HPLC measurements. Therefore, the predicted Casatellite measurements should naturally lean toward the fluorometric values. However, this is not what we found, suggesting that the species composition in the Western Antarctic Peninsula region may be different from the “mean” composition in the original algorithm dataset.

The large difference observed between CaFluor and CaHPLC from the same water samples was likely due, in part, to interference of the fluorescence signal by chlorophyll accessory pigments (Cb,Ccandtheirdegradationproducts). In our studyCbonly occurred in low concentrations compared to Ca(mean ratio Cb/Ca = 0.023, n= 486); however, Ccwas relatively high (mean ratio Cc/Ca = 0.25, n= 486) (Fig. 8). The presence of significant amounts of Cc is known to cause an overestimation of Ca by the fluorometric method (Gibbs, 1979; Lorenzen, 1981).

The presence of significant amounts of Cc appears to be the cause for the overestimation of Ca by the fluorometric method.

Cb is an accessory pigment in prochlorophytes, chlorophytes and prasinophytes, while Ccis generally present in diatoms, dinoflagellates, cryptophytes and haptophytes (Parsons et al., 1984). Diatoms are the dominant phytoplankton in waters west of the Antarctic Peninsula, with dinoflagellates being very abundant at times (Prézelin et al, 2000, 2004). Prochlorophytes, a type of cyanobacteria first identified in the late 1980s (Chisholm et al., 1988), have not yet been identified observed in the Southern Ocean, while chlorophytes can be abundant (Prézelin et al, 2000, 2004). Similarly, cryptophytes are usually scarce in the water column, but can be very abundant in coastal surface melt water during spring and summer (Moline and Prézelin, 1996). Alloxanthin, the biomarker pigment for cryptophytes (Prézelin et al, 2000), occurred in 91% (n=516) of the pigment samples. Hence, chlorophytes were probably the dominant source of Cb during our study period, whilethe dominant sources of Cc appear to be diatoms, dinoflagellates and cryptophytes,identified by the presence of fucoxanthin, peridinin, and alloxanthin in 99.5%, 53% and 91% of the samples, respectively.